Re: [PATCH v8 3/6] clocksource: Ingenic: Add high resolution timer support for SMP.

From: Paul Cercueil
Date: Tue May 19 2020 - 13:43:19 EST

Next message: Mark Brown: "Re: [PATCH 3/3] selftests: vdso: Add a selftest for vDSO getcpu()"
Previous message: Mark Brown: "Re: [PATCH 2/3] selftests: vdso: Use a header file to prototype parse_vdso API"
In reply to: åçæ (Zhou Yanjie): "[PATCH v8 3/6] clocksource: Ingenic: Add high resolution timer support for SMP."
Next in thread: Paul Cercueil: "[PATCH] clocksource: Ingenic: Add high resolution timer support for SMP."
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Hi Zhou,

Le mar. 19 mai 2020 Ã 22:35, åçæ (Zhou Yanjie) <zhouyanjie@xxxxxxxxxxxxxx> a Ãcrit :

Enable clock event handling on per CPU core basis.
Make sure that interrupts raised on the first core execute
event handlers on the correct CPU core.

Tested-by: H. Nikolaus Schaller <hns@xxxxxxxxxxxxx>
Tested-by: Paul Boddie <paul@xxxxxxxxxxxxx>
Signed-off-by: åçæ (Zhou Yanjie) <zhouyanjie@xxxxxxxxxxxxxx>
---

Notes:
v1->v2:
1.Adjust function naming to make it more reasonable.
2.Replace function smp_call_function_single() with
smp_call_function_single_async() in order to resolve
the warning below:

[ 0.350942] smp: Brought up 1 node, 2 CPUs
[ 0.365497] ------------[ cut here ]------------
[ 0.365522] WARNING: CPU: 0 PID: 1 at kernel/smp.c:300 smp_call_function_single+0x110/0x200
[ 0.365533] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.5.0-rc1+ #5
[ 0.365537] Stack : 00000000 59c73bcd 00000037 80074e80 80000000 80670000 805a0000 80620590
[ 0.365557] 8065ce38 8fc0dc8c 806d0000 00000000 80670000 00000001 8fc0dc20 59c73bcd
[ 0.365574] 00000000 00000000 806f0000 80670000 00000000 806dab00 00000000 2d302e35
[ 0.365591] 203a6d6d 806e0000 806e0000 70617773 80670000 00000000 00000000 00000009
[ 0.365610] 00000000 8fc94e20 8fc0de30 80690000 00000018 803592dc 00000000 806d0000
[ 0.365627] ...
[ 0.365634] Call Trace:
[ 0.365647] [<8001b9a0>] show_stack+0x6c/0x12c
[ 0.365663] [<804aed20>] dump_stack+0x98/0xc8
[ 0.365673] [<8003044c>] __warn+0xc4/0xe8
[ 0.365682] [<800304f4>] warn_slowpath_fmt+0x84/0xb8
[ 0.365690] [<800a886c>] smp_call_function_single+0x110/0x200
[ 0.365703] ---[ end trace 5785856ca39c79d5 ]---
[ 0.365557] 8065ce38 8fc0dc8c 806d0000 00000000 80670000 00000001 8fc0dc20 59c73bcd
[ 0.365574] 00000000 00000000 806f0000 80670000 00000000 806dab00 00000000 2d302e35
[ 0.365591] 203a6d6d 806e0000 806e0000 70617773 80670000 00000000 00000000 00000009
[ 0.365610] 00000000 8fc94e20 8fc0de30 80690000 00000018 803592dc 00000000 806d0000
[ 0.365627] ...
[ 0.365634] Call Trace:
[ 0.365647] [<8001b9a0>] show_stack+0x6c/0x12c
[ 0.365663] [<804aed20>] dump_stack+0x98/0xc8
[ 0.365673] [<8003044c>] __warn+0xc4/0xe8
[ 0.365682] [<800304f4>] warn_slowpath_fmt+0x84/0xb8
[ 0.365690] [<800a886c>] smp_call_function_single+0x110/0x200
[ 0.365703] ---[ end trace 5785856ca39c79d5 ]---

v2->v3:
No Change.

v3->v4:
Rebase on top of kernel 5.6-rc1.

v4->v5:
Move the check for (evt->event_handler) from "ingenic_per_cpu_event_handler"
to "ingenic_tcu_cevt_cb".

v5->v6:
No change.

v6->v7:
Remove unnecessary check for "NR_CPUS > 1".

v7->v8:
Use "num_possible_cpus()" instead "NR_CPUS".
Reported-by: kbuild test robot <lkp@xxxxxxxxx>

drivers/clocksource/ingenic-timer.c | 103 ++++++++++++++++++++++++++++--------
1 file changed, 82 insertions(+), 21 deletions(-)

diff --git a/drivers/clocksource/ingenic-timer.c b/drivers/clocksource/ingenic-timer.c
index 4963336..230e996 100644
--- a/drivers/clocksource/ingenic-timer.c
+++ b/drivers/clocksource/ingenic-timer.c
@@ -1,7 +1,8 @@
// SPDX-License-Identifier: GPL-2.0
/*
- * JZ47xx SoCs TCU IRQ driver
+ * XBurst SoCs TCU IRQ driver

If you want to get rid of the JZ47xx, then just write 'Ingenic SoCs TCU IRQ driver', since XBurst is only the name of the CPU. Also, this belongs to a separate patch.

* Copyright (C) 2019 Paul Cercueil <paul@xxxxxxxxxxxxxxx>
+ * Copyright (C) 2020 åçæ (Zhou Yanjie) <zhouyanjie@xxxxxxxxxxxxxx>
*/

#include <linux/bitops.h>
@@ -21,18 +22,23 @@

#include <dt-bindings/clock/ingenic,tcu.h>

+static DEFINE_PER_CPU(call_single_data_t, ingenic_cevt_csd);
+
struct ingenic_soc_info {
unsigned int num_channels;
};

struct ingenic_tcu {
struct regmap *map;
+ struct device_node *np;
struct clk *timer_clk, *cs_clk;
+ unsigned int timer_local[NR_CPUS];

NR_CPUS can be very big, so this would make the struct ingenic_tcu very large. What you could do, is have a variable array at the end of the struct:

unsigned int timer_local[];

Then you can create the ingenic_tcu struct using struct_size() from <linux/overflow.h>:

tcu = kzalloc(struct_size(tcu, timer_local, num_possible_cpus()), GFP_KERNEL);

-Paul

unsigned int timer_channel, cs_channel;
struct clock_event_device cevt;
struct clocksource cs;
- char name[4];
+ char name[8];
unsigned long pwm_channels_mask;
+ int cpu;
};

static struct ingenic_tcu *ingenic_tcu;
@@ -81,6 +87,24 @@ static int ingenic_tcu_cevt_set_next(unsigned long next,
return 0;
}

+static void ingenic_per_cpu_event_handler(void *info)
+{
+ struct clock_event_device *cevt = (struct clock_event_device *) info;
+
+ cevt->event_handler(cevt);
+}
+
+static void ingenic_tcu_per_cpu_cb(struct clock_event_device *evt)
+{
+ struct ingenic_tcu *tcu = to_ingenic_tcu(evt);
+ call_single_data_t *csd;
+
+ csd = &per_cpu(ingenic_cevt_csd, tcu->cpu);
+ csd->info = (void *) evt;
+ csd->func = ingenic_per_cpu_event_handler;
+ smp_call_function_single_async(tcu->cpu, csd);
+}
+
static irqreturn_t ingenic_tcu_cevt_cb(int irq, void *dev_id)
{
struct clock_event_device *evt = dev_id;
@@ -89,7 +113,7 @@ static irqreturn_t ingenic_tcu_cevt_cb(int irq, void *dev_id)
regmap_write(tcu->map, TCU_REG_TECR, BIT(tcu->timer_channel));

if (evt->event_handler)
- evt->event_handler(evt);
+ ingenic_tcu_per_cpu_cb(evt);

return IRQ_HANDLED;
}
@@ -105,14 +129,21 @@ static struct clk * __init ingenic_tcu_get_clock(struct device_node *np, int id)
return of_clk_get_from_provider(&args);
}

-static int __init ingenic_tcu_timer_init(struct device_node *np,
- struct ingenic_tcu *tcu)
+static int ingenic_tcu_setup_per_cpu_cevt(struct device_node *np,
+ unsigned int channel)
{
- unsigned int timer_virq, channel = tcu->timer_channel;
+ unsigned int timer_virq;
struct irq_domain *domain;
+ struct ingenic_tcu *tcu;
unsigned long rate;
int err;

+ tcu = kzalloc(sizeof(*tcu), GFP_KERNEL);
+ if (!tcu)
+ return -ENOMEM;

I have no idea why you're doing that, but this is really wrong and highly confusing. The ingenic_tcu instance is already created in the probe. If you need per-timer data, then create another structure.

-Paul

+
+ tcu->map = ingenic_tcu->map;
+
tcu->timer_clk = ingenic_tcu_get_clock(np, channel);
if (IS_ERR(tcu->timer_clk))
return PTR_ERR(tcu->timer_clk);
@@ -139,13 +170,15 @@ static int __init ingenic_tcu_timer_init(struct device_node *np,
goto err_clk_disable;
}

- snprintf(tcu->name, sizeof(tcu->name), "TCU");
+ snprintf(tcu->name, sizeof(tcu->name), "TCU%u", channel);

err = request_irq(timer_virq, ingenic_tcu_cevt_cb, IRQF_TIMER,
tcu->name, &tcu->cevt);
if (err)
goto err_irq_dispose_mapping;

+ tcu->cpu = smp_processor_id();
+ tcu->timer_channel = channel;
tcu->cevt.cpumask = cpumask_of(smp_processor_id());
tcu->cevt.features = CLOCK_EVT_FEAT_ONESHOT;
tcu->cevt.name = tcu->name;
@@ -166,6 +199,25 @@ static int __init ingenic_tcu_timer_init(struct device_node *np,
return err;
}

+static int ingenic_tcu_setup_cevt(unsigned int cpu)
+{
+ int ret;
+
+ ret = ingenic_tcu_setup_per_cpu_cevt(ingenic_tcu->np,
+ ingenic_tcu->timer_local[cpu]);
+ if (ret)
+ goto err_tcu_clocksource_cleanup;
+
+ return 0;
+
+err_tcu_clocksource_cleanup:
+ clocksource_unregister(&ingenic_tcu->cs);
+ clk_disable_unprepare(ingenic_tcu->cs_clk);
+ clk_put(ingenic_tcu->cs_clk);
+ kfree(ingenic_tcu);
+ return ret;
+}
+
static int __init ingenic_tcu_clocksource_init(struct device_node *np,
struct ingenic_tcu *tcu)
{
@@ -240,6 +292,7 @@ static int __init ingenic_tcu_init(struct device_node *np)
const struct ingenic_soc_info *soc_info = id->data;
struct ingenic_tcu *tcu;
struct regmap *map;
+ unsigned cpu = 0;
long rate;
int ret;

@@ -253,13 +306,18 @@ static int __init ingenic_tcu_init(struct device_node *np)
if (!tcu)
return -ENOMEM;

- /* Enable all TCU channels for PWM use by default except channels 0/1 */
- tcu->pwm_channels_mask = GENMASK(soc_info->num_channels - 1, 2);
+ /*
+ * Enable all TCU channels for PWM use by default except channels 0/1,
+ * and channel 2 if target CPU is JZ4780 and SMP is selected.
+ */
+ tcu->pwm_channels_mask = GENMASK(soc_info->num_channels - 1,
+ num_possible_cpus() + 1);
of_property_read_u32(np, "ingenic,pwm-channels-mask",
(u32 *)&tcu->pwm_channels_mask);

- /* Verify that we have at least two free channels */
- if (hweight8(tcu->pwm_channels_mask) > soc_info->num_channels - 2) {
+ /* Verify that we have at least num_possible_cpus() + 1 free channels */
+ if (hweight8(tcu->pwm_channels_mask) >
+ soc_info->num_channels - num_possible_cpus() + 1) {
pr_crit("%s: Invalid PWM channel mask: 0x%02lx\n", __func__,
tcu->pwm_channels_mask);
ret = -EINVAL;
@@ -267,13 +325,19 @@ static int __init ingenic_tcu_init(struct device_node *np)
}

tcu->map = map;
+ tcu->np = np;
ingenic_tcu = tcu;

- tcu->timer_channel = find_first_zero_bit(&tcu->pwm_channels_mask,
+ tcu->timer_local[cpu] = find_first_zero_bit(&tcu->pwm_channels_mask,
soc_info->num_channels);
+
+ for (cpu = 1; cpu < num_possible_cpus(); cpu++)
+ tcu->timer_local[cpu] = find_next_zero_bit(
+ &tcu->pwm_channels_mask, soc_info->num_channels,
+ tcu->timer_local[cpu - 1] + 1);
+
tcu->cs_channel = find_next_zero_bit(&tcu->pwm_channels_mask,
- soc_info->num_channels,
- tcu->timer_channel + 1);
+ soc_info->num_channels, tcu->timer_local[cpu - 1] + 1);

ret = ingenic_tcu_clocksource_init(np, tcu);
if (ret) {
@@ -281,9 +345,10 @@ static int __init ingenic_tcu_init(struct device_node *np)
goto err_free_ingenic_tcu;
}

- ret = ingenic_tcu_timer_init(np, tcu);
- if (ret)
- goto err_tcu_clocksource_cleanup;
+ /* Setup clock events on each CPU core */
+ ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "Ingenic XBurst: online",
+ ingenic_tcu_setup_cevt, NULL);
+ WARN_ON(ret < 0);

/* Register the sched_clock at the end as there's no way to undo it */
rate = clk_get_rate(tcu->cs_clk);
@@ -291,10 +356,6 @@ static int __init ingenic_tcu_init(struct device_node *np)

return 0;

-err_tcu_clocksource_cleanup:
- clocksource_unregister(&tcu->cs);
- clk_disable_unprepare(tcu->cs_clk);
- clk_put(tcu->cs_clk);
err_free_ingenic_tcu:
kfree(tcu);
return ret;
--
2.7.4

Next message: Mark Brown: "Re: [PATCH 3/3] selftests: vdso: Add a selftest for vDSO getcpu()"
Previous message: Mark Brown: "Re: [PATCH 2/3] selftests: vdso: Use a header file to prototype parse_vdso API"
In reply to: åçæ (Zhou Yanjie): "[PATCH v8 3/6] clocksource: Ingenic: Add high resolution timer support for SMP."
Next in thread: Paul Cercueil: "[PATCH] clocksource: Ingenic: Add high resolution timer support for SMP."
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]